Setting and data sources

The study was a nationwide Danish register-based cohort study. In Denmark, every citizen is registered with a unique 10-digit civil personal registration (CPR) number, and we used this to link information at the individual level between the following national registries. The Danish National Patient Registry, which contains nationwide information on primary and secondary diagnoses associated with each inpatient, outpatient, and emergency room visit according to the International Classification of Diseases, 10th version (ICD-10) [30]. The Danish Register of Medicinal Product Statistics, which contains information on redeemed prescriptions of medication in community pharmacies. The information includes type and product name of medication [31,32]. The Danish National Health Insurance Service Registry, which contains information on contacts and specific services provided by the general practitioner (GP). The register is a reimbursement register and provides no information on reasons for contact, diagnosis, or test results [27,28]. Statistics Denmark, which contains information on population statistics including socioeconomic variables. Electronic Patient Journals, to identify females referred to an HPV centre for suspected adverse effects. There is no study protocol available for publication.

Study population

The study population included all HPV-vaccinated females living in Denmark, born between 1974 and 2006, and vaccinated in the period between January 1, 2006 and December 31, 2017. Females who received the first HPV vaccination at the GP were identified in the Danish National Health Insurance Service Registry, in which the week of HPV vaccination is registered (service codes 8328, 8329, 8330 or 8334, 8335, and 8336). In addition, females who redeemed a first prescription of a HPV vaccine (4-valent, 2-valent, or 9-valent) at a community pharmacy were identified in the Danish Register of Medicinal Product Statistics (Anatomic Therapeutic Chemical (ATC) codes J07BM01, J07MB02, or J07BM03). The HPV vaccination date was a random allocated date in the week of receiving the first HPV vaccination at the GP or the date of redeeming the first HPV vaccine prescription at the pharmacy. If a female was registered with both a redemption of the HPV vaccine and a service code of HPV vaccination at the GP within a period of 14 days, we used the first date registered.

Exposure

The exposure of interest was having an infection in the period from 1 month before until 1 month after the first HPV vaccination date. Presence of infection was defined by the following markers: (1) admission date for an inpatient or outpatient hospital contact associated with a primary (first-listed) discharge diagnosis of infection (specific ICD-10 codes in S1 Table); (2) date of redeeming at least one prescription of an anti-infective medication at a pharmacy (specific ATC codes in S2 and S3 Tables); and (3) date of a rapid streptococcal test (service code 7109) used to assist in the diagnosis of suspected pharyngitis at the GP. We categorised and included each female at the date of the most severe infection category ± 30 days around the HPV vaccination, with severity going from hospital-treated infection (most severe), through redemption of anti-infective medication only, and to rapid streptococcal test for pharyngitis without anti-infectives or hospital contact (least severe). In subanalyses, we further categorised anti-infective medications according to type as an indication of likely pathogens treated: bacterial, viral, fungal infection, and multiple infections (specific ATC codes in S2 Table), and according to likely site of the treated infection, e.g., urinary or upper respiratory tract (specific ATC codes in S3 Table). In a subanalysis, we excluded prescriptions of tetracycline, as these can be prescribed for several conditions including treatment of severe acne.

Outcome

Females could be referred to any of the HPV centres, established in 2015, by their GP or by a medical specialist. The criteria for referral were unspecific or unexplained symptoms occurring in temporal relation to HPV vaccination. The outcome in this study was referral to one of these centres at some point in the period from July 1, 2015 until December 31, 2017.

Confounding factors

Variables that may potentially confound the associations by being associated with both risk of infection and risk of later suspected adverse vaccine effects were identified a priori. Information related to social position in terms of maternal education and socioeconomic status of the family was obtained from Statistics Denmark for the year before first HPV vaccination [33]. We obtained information on chronic somatic conditions from the Danish National Patient Registry. Information on psychiatric conditions and asthma was obtained from the Danish National Patient Registry as well as from the Danish Register of Medicinal Product Statistics (see specific ICD-10 codes and ATC codes in S1 Text) [28,29]. Information on the somatic and psychiatric conditions were obtained in a period of 5 years before the date of first HPV vaccination. Finally, we accounted for age and calendar year as use of antibiotics decreased from 2008 to 2017 in all young age groups in Denmark [34].

Statistical analyses

Characteristics of the study population according to infection in temporal proximity to HPV vaccination status are presented using number and percentages. Also, we graphically present the age distribution of the HPV vaccinated females according to year of vaccination as several catch-up programmes including older girls and young women have taken place in different years. All models were defined prior to conducting the study. We used simple and multivariable logistic regression to estimate the crude and adjusted odds ratio (OR) with 95% confidence intervals (CIs) as a measure of the association between each specific definition of infection and referral to an HPV centre. In a sensitivity analysis, log binomial regression analysis was applied instead. The associations were estimated both overall and stratified on age at vaccination (±18 years) and year of vaccination (2006 to 2012 and 2013 to 2017), respectively. All analyses were adjusted for the following variables: year of vaccination (2006 to 2008, 2009 to 2011, 2012 to 2013, 2014 to 2015, 2016 to 2017), age at vaccination (11 to 14, 15 to 17, 18 to 24, ≥25 years old), maternal educational level (PhD, higher education, medium-length education, basic education, high school or vocational, and primary school), socioeconomic status of the family (owner of business, chief executive or employee with high income, employee with middle income, employee with low income, employee with unspecified income, unemployed, and pensioner), as well as chronic somatic and psychiatric conditions in the 5 years prior to the first vaccine registration (yes/no). Since referral to an HPV centre is rare and we have virtually complete follow-up in the study period, the calculated ORs in the cohort study were interpreted as risk ratios.

Case-crossover and case time-control analyses

In order to understand whether unmeasured confounding could explain potential associations between infections in temporal proximity to vaccination and referral to an HPV centre, we performed a case-crossover and a case time-control design (Fig 1) [35,36]. In a first step, we used a case-crossover design, which only includes cases (all females referred to an HPV centre) and uses cases as their own controls (using different exposure periods). Thus, this design can effectively control for all personal time-stable factors. For each case, each type of infection was ascertained for the same “risk-window period” as the main analysis (±1 month of vaccination) and a “control-window period” of the same length the year before (11 to 13 months before vaccination). Exposure frequencies from the risk-window period and the control-window period were compared by calculating a matched OR. The case-crossover OR (OR_cases) was estimated as the ratio of the number of cases exposed only in the risk-window period to the number of cases exposed only during the control-window period (i.e., the ratio of discordant pairs). In a second step, we used a case time-control design in order to control for temporal changes in infections from 1 year to the other. To do that, we estimated the matched OR for controls (all nonreferred females) as we did for cases: For each control, exposure frequencies from risk-window period and the control-window period were compared through an OR (OR_controls). The case time-control OR was estimated as OR_adjusted = OR_cases / OR_controls. All p-values related to specific estimates in the statistical analyses were estimated by Wald tests.

All statistical analyses were carried out using STATA 13 statistical software.

This study is reported as per the RECORD-PE guideline (S1 Checklist).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Case-crossover and case time-control design (inspired by Hernández-Díaz and colleagues [33]).

https://doi.org/10.1371/journal.pmed.1003768.g001

Case-crossover and case time-control analyses

Although controls were more likely to experience a hospital-treated infection or anti-infective prescription in temporal proximity to vaccination than in the year before (indicating moderate trends of exposure to an infection), cases were even more likely to have an infection around vaccination than in the year before (Table 4). Thus, after adjustment for the time trend, we observed an association between anti-infective medication redeemed around the time of HPV vaccination and later referral to an HPV centre (OR 1.40, 95% CI 1.08 to 1.81; P = 0.011). Although the association with referral to an HPV centre was based on a small number of discordant pairs and not statistically significant for a hospital-treated infection happening around the time of HPV vaccination, the point estimate was larger than for those with anti-infective prescriptions (OR 1.84, 95% CI 0.74 to 4.56; P = 0.189). On the contrary, the association for rapid streptococcal test at the GP and referral to an HPV centre due to suspected adverse effects was weaker (OR 1.11, 95% CI 0.73 to 1.70; P = 0.718).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Case-crossover and case time-control analyses of the association between infection in temporal proximity to HPV vaccination (±1 month) and referral to an HPV centre (reference period 11–13 months before HPV vaccination).

https://doi.org/10.1371/journal.pmed.1003768.t004

Strengths and limitations

A strength of our study is the population-based design with virtually complete follow-up, which limits the risk of selection bias. Furthermore, in Denmark, there is free access to both primary and secondary healthcare, and all information in the registries on diagnoses, prescriptions, and special services were collected prospectively. Generally, the information from The Danish National Patient Registry and Danish Register of Medicinal Product Statistics are of high quality, and the validity of diagnoses and medication is high [30–32], and we therefore consider the risk of information bias for these exposures low. Information on rapid streptococcal test from the Danish National Health Insurance Service Registry does not include any information on diagnoses or positive/negative test results and is therefore mainly a measure of indication/suspicion of an infection. The GPs’ decision to provide a test or not in case of infection symptoms in temporal proximity to HPV vaccination is most likely unrelated to later referral to an HPV centre, but the females’ individual healthcare attendance behaviour might influence the likelihood of having a test for infection, and some infections might not be captured. Also, females with later symptoms who could recall other health effects (e.g., infection) happening around the time of HPV vaccination might in theory be less likely to be referred, as they had an alternative explanation for their symptoms. However, this potential detection bias cannot explain the positive associations found in the present study. Another strength in our study is that we were able to adjust for a range of confounders, including prior somatic conditions, asthma, and psychiatric conditions, based on highly valid information from registries. Additionally, we conducted a case-crossover and a case time-control analyses in order to control for unmeasured/uncontrolled confounding including potential differences in healthcare attendance pattern. The results of the case-crossover and case time-control analyses supported our findings for females with hospital-treated infections and for females who had redeemed anti-infective medication. Conversely, the analysis did not support the finding for females tested for pharyngitis at the GP. A possible explanation for this could be that having a test for infection at the GP is more likely influenced by healthcare attendance pattern than a hospital-treated infection or receiving medication for an infection. This explanation is supported by Danish case–control studies reporting differences in healthcare attendance prior to HPV vaccination between females referred to an HPV centre and nonreferred females [37–39]. The associations detected for the treated infections using both study designs support the robustness of these findings. A limitation of our study is that the HPV centres did not open until June 2015. We could therefore not include time to referral in our analyses. Also, some females experiencing adverse effects well before this date might have had other diagnoses and treatment and consequently never been referred to an HPV centre. To assess any bias from possible incomplete outcome registration early in the study period, we stratified the main analysis on calendar year of first HPV vaccination (2006 to 2012 and 2013 to 2017; see S4 Table). No major differences in relative risk estimates across the periods were observed.

Comparison with other studies

To our knowledge, no other studies have investigated the association between infections in temporal proximity to HPV vaccination and suspected adverse vaccine effects. However, some studies have investigated the association between infection or vaccination and the CFS diagnosis. In a Norwegian nationwide register-based follow-up study, the researchers found that CFS was associated with pandemic influenza infection, but not with a pandemic influenza vaccine [26]. Correspondingly, in a case–control study, no association between vaccination against meningococcal disease in teenagers and CFS was observed, whereas the authors found an association between infectious mononucleosis and CFS [25]. Finally, no association between HPV vaccination and CFS was detected in another Norwegian nationwide register-based follow-up study [8]. These studies support our hypothesis of infection in temporal proximity to HPV vaccination as a potential trigger of the CFS-like symptoms. As we only included HPV-vaccinated females in our study, we are not able to differentiate possible effects of an infection in itself from a potential interaction between an infection and the HPV vaccination. Furthermore, we are not able to differentiate effects of the infection per se from the treatment given for the infection. Future studies should aim at investigating the specific mechanisms further.